7 Cents a Second: The Star of Google's Video Model This Time Is the Bill

Google has released Gemini Omni 1.2 Flash. The official list of updates is very specific: lower cost, video generation and editing length limit raised to 15 seconds, improved temporal coherence of scenes, upgraded character voices, fewer visual artifacts such as jump cuts, and higher fidelity to uploaded reference materials.

The official demo used a single style frame to generate a 4-shot, 15-second educational short about a beehive. The camera cuts four times, but the bees are still the same batch of bees.

What's worth watching in this demo is not how exquisite the visuals are, but consistency. Objects don't drift between multiple shots, and characters don't change faces—areas where previous models often failed.

Price is another focus this time. 720p video on the API costs $0.07 per second, about $1.04 for 15 seconds, a 30% drop from the previous generation. Twitter user 0xBaron added the detailed math: 720p is $12 per million video tokens, equivalent to $0.07 per second; calls to version 1.1 are billed at the 1.2 price starting December 3; 1080p upsampling is $0.104 per second.

One user mentioned a comparison: last October, you still had to write a 9,500-character brief and spend 12 hours of compute to get a barely coherent clip; now it's one dollar, 15 seconds, and the character is still the same person. His conclusion: 'The price is the announcement.'

Another user's judgment is more direct: this pricing is for marketers to test ad variants at scale, not for filmmakers. Conversely, $0.07 per second doesn't determine how good a single video can be, but how many times a team can try in a day.

On the hands-on side, a user posted a video they generated with Omni, commenting that it was 'unexpectedly good.' Some in the replies also mentioned that the previous voice issue was 'American-style Chinese,' which has improved this time.

Questions haven't stopped: some ask whether it can run on local hardware, some say Flow currently doesn't support using Omni for extending, and others hope the input can support audio plus images. These are not small issues, but when placed alongside 15 seconds and one dollar, their priority seems less high.

There are several entry points: scene directing and multi-turn editing in Flow, free generation of 15-second clips in Google Vids, 15-second video generation in the Gemini App, and API access for developers.

15 seconds, one dollar. If this price point holds, how AI video is used will change: from 'generate once to see the effect' to 'try in batches and pick one that works.'